Every repo I finish ends the same way. The code works. The tests pass. Then I stare at a blank README wondering where to even start. So I built something to fix that. Meet scriba. Paste any public GitHub repo link and it reads the actual code, the real files, not a generic template, and writes back structured documentation in under thirty seconds. Overview, tech stack, setup steps, key modules, everything you'd actually want to know before touching the code. It's free, needs no signup, and only touches public repos. I built it to show what an AI agent can really do when it's reasoning over a live codebase instead of just chatting back and forth. Now I want to break it. Paste your messiest, weirdest repo and tell me what it gets wrong. Try it here: scriba-orpin.vercel.app (http://scriba-orpin.vercel.app) What repo are you testing it on?

1

22

Rhythm: An AI daily planner that plans when, not just what I kept missing my own basic goals. Drink water, stretch, read for twenty minutes: I had no trouble writing them down, I just never had a place to put them. Every habit app I tried handed me a list and left the scheduling to me, and "later" always lost to whatever the day actually threw at me. So I built the tool I actually needed. Rhythm takes the fixed shape of your day (wake time, sleep time, work hours) and slots your goals into the real open windows around it, using an AI scheduling engine that explains its reasoning for every placement instead of acting as a black box. It handles three distinct kinds of goals: simple recurring habits, several-times-a-day reminders (spaced automatically, tightened on the fly if you skip one), and bigger goals broken into a paced, multi-week plan toward a deadline you set. Built with Next.js, React, and a real OpenAI-backed scheduling API, with a calm, warm, journal-like design system, hand-tuned color and type tokens, organic shapes instead of generic dashboard chrome, and zero gamified streak badges. Every screen, from onboarding to the reasoning card behind each task, was designed to feel like a planner, not a productivity dashboard. This is the first pass, an early design concept proving out the core idea. What's here is a working web prototype; what's next is a proper native mobile app with real push notifications, calendar sync, and a fuller feature set built on top of this foundation.

5

4

241

Things I believe after building with AI: - If you do something manually more than twice, automate it. - AI agents are the most fun thing you can build right now. - A demo that works beats a pitch deck every time. - The best engineering is invisible. You only notice when it's missing. Hi, I'm Ayesha, a Full-stack developer in Berlin. I spend my days in TypeScript, React and Python, mostly teaching LLMs to do real work. Think agents that plan code changes and push them to GitHub on their own, RAG assistants, CRM automation that cut response times in half. I'm new here on Contra so I put together a quick tour of my portfolio and what I build. My favorite part is the AI agent that runs a Kanban board by itself. Full breakdown coming in my next post. Your turn. What's rule #1 in your work?

2

3

1.2K

Cover image for Anka Diversify — AI-Native Project

Anka Diversify — AI-Native Project Management Platform A full-stack SaaS platform built for software teams, with AI woven into the core workflow rather than bolted on as a feature. The centerpiece is an autonomous AI coding agent built on the Anthropic and OpenAI APIs using tool-use and multi-step planning. Give it a task, and it plans the necessary code changes across multiple files, generates the implementation, presents a reviewable diff, and pushes directly to GitHub once approved, handling multi-file commits properly via the GitHub API's tree and blob objects. Getting this to work reliably in production, not just in demos, was the real engineering challenge. LLM output isn't always clean: malformed JSON, inconsistent formatting, file references that don't match the actual repo. I built a validation and retry layer that checks structure and re-prompts with corrective context on failure, plus logging to catch failure patterns before they hit users. Beyond the coding agent, the platform includes a project-scoped AI assistant with repo context injection (a lightweight RAG pattern), real-time team chat, Kanban boards with drag-and-drop, sprint planning, S3 file uploads, and role-based access control. Stack: Next.js 16, React 19, TypeScript, Tailwind CSS, Shadcn/UI, Express 5, Prisma 5, PostgreSQL, Anthropic API, OpenAI API Live: anka-diversify-os-production.up.railway.app (http://anka-diversify-os-production.up.railway.app)

0

157

Ayesha Rashid Khan

AI Developer: Power Your Business with AI